Citation: Maktab Dar Oghaz, M.;
Razaak, M.; Remagnino, P. Enhanced
Single Shot Small Object Detector
for Aerial Imagery Using
Super-Resolution, Feature Fusion and
Deconvolution. Sensors 2022, 22, 4339.
https://doi.org/10.3390/s22124339
Academic Editors: Jaroslaw Pytka,
Andrzej Łukaszewicz, Zbigniew
Kulesza, Wojciech Giernacki, Andriy
Holovatyy and Petros Daras
Received: 21 March 2022
Accepted: 27 May 2022
Published: 8 June 2022
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2022 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
Article
Enhanced Single Shot Small Object Detector for Aerial Imagery
Using Super-Resolution, Feature Fusion and Deconvolution
Mahdi Maktab Dar Oghaz
1,
*
,†
, Manzoor Razaak
2,†
and Paolo Remagnino
3,†
1
Faculty of Science and Engineering, Anglia Ruskin University, Cambridge CB1 1PT, UK
2
The Robot Vision Team, Kingston University London, London KT1 2EE, UK; manzoor.razak@kingston.ac.uk
3
Department of Computer Science, Durham University, Upper MountJoy, Durham DH1 3LE, UK;
paolo.remagnino@gmail.com
* Correspondence: mahdi.maktabdar@aru.ac.uk
† These authors contributed equally to this work.
Abstract:
One common issue of object detection in aerial imagery is the small size of objects in pro-
portion to the overall image size. This is mainly caused by high camera altitude and wide-angle
lenses that are commonly used in drones aimed to maximize the coverage. State-of-the-art general
purpose object detector tend to under-perform and struggle with small object detection due to loss
of spatial features and weak feature representation of the small objects and sheer imbalance between
objects and the background. This paper aims to address small object detection in aerial imagery by
offering a Convolutional Neural Network (CNN) model that utilizes the Single Shot multi-box Detec-
tor (SSD) as the baseline network and extends its small object detection performance with feature
enhancement modules including super-resolution, deconvolution and feature fusion. These modules
are collectively aimed at improving the feature representation of small objects at the prediction layer.
The performance of the proposed model is evaluated using three datasets including two aerial images
datasets that mainly consist of small objects. The proposed model is compared with the state-of-
the-art small object detectors. Experiment results demonstrate improvements in the mean Absolute
Precision (mAP) and Recall values in comparison to the state-of-the-art small object detectors that
investigated in this study.
Keywords: deconvolution; feature fusion; small object detection; SSD; super-resolution
1. Introduction
Object detection is one of the core research areas in computer vision. Recent break-
throughs in Convolutional Neural Network (CNN) and object detection unlocked new
horizons and possibilities in various domains ranging from security and surveillance appli-
cations, such as face detection, crowd analysis and activity recognition to medical image
analysis and self-driving vehicles research [1–4].
Despite the contextual similarities of these domains, they utilize different image
acquisition techniques that often require significant adaptation and alteration of the state-of-
the-art general purpose object detectors to achieve desirable results. A prominent example
of such a domain is Unmanned Aerial Vehicles (UAV) imagery. The UAV imagery is getting
more popular than ever before with a variety of applications including smart farming [
5
],
search and rescue [
6
], disaster management [
7
], archaeological structure modeling [
8
],
security and surveillance [
9
] and many others. In UAV imagery, due to the flight altitude,
the top-down camera perspective and wide-angle lenses, object shapes and appearances
are relatively unconventional and they usually take up a small fraction of the image area,
as illustrated in Figure 1. General purpose object detectors are trained and tuned on datasets,
such as ImageNet and COCO, which mainly offer ground-level medium-sized images.
These detectors fail to provide good detection accuracy when it comes to out-of-ordinary
Sensors 2022, 22, 4339. https://doi.org/10.3390/s22124339 https://www.mdpi.com/journal/sensors
